GeneMark-ES is a self-training algorithm and often does not work with small genomes. It needs at least 10 Mb for training only. I would suggest using GeneMark.hmm (eukaryotic).
This error is result of "probuild" function of genemark, so it could be because of non-canonical base-characters or sequences are just too small. You can also try Augustus if you prefer.
It really is an error of "probuild". Lower case characters (figured it out just some minutes ago). But thanks for the answer. My set of scaffolds has 31 Mb,
what do you mean with "lower case characters"? I attempted to run with inputs of sizes 13Mb and 600Mb and I got the error: error, file not found: info/training.fna. I checked in a few forums (https://www.biostars.org/p/262992/ , https://github.com/nextgenusfs/funannotate/issues/150) and they point that the problem is the probuild function. My input data seems to be okay, but I wonder what else could be the problem.
Some bases in my contigs were in lower case. I don't recall what version of Genemark-ES I was using that time. When I changed the lower case bases for upper case bases the program executed just fine.